Creativity in the Era of Artificial Intelligence¶
Authors: Philippe Esling, Ninon Devis
The current focus on mimicking human capabilities at the intersection of creativity and AI is counterproductive and an underutilization of the potential that AI has to offer. This paper details the various aspects of creativity from a value-creation and process perspective, highlighting where AI might be a useful mechanism to augment rather than replace our abilities.
If you enjoy the content on this page, you can support my work by buying me a coffee
When we think of AI-based creativity today, most of our attention is drawn to the GAN-style works that are a mash-up of existing creative works be those visual, literary, or auditory. But, through an examination of the epistemology of creativity, the paper illustrates that we are over-indexing on only certain aspects of creativity while ignoring others. The paper talks about how we can better harness the potential of AI to expand the meaning of creativity and create experiences and art that is beyond the limitations that humans impose on them. This also includes the limits inherent to the current crop of AI technologies in how they are designed. An analysis of these limitations reveals that perhaps co-creativity is what we need to be aiming for and move from the notion of artificial intelligence in creativity to artificial creativity.
Epistemology of creativity¶
Creative work has significance not just because of its content but also because of its positioning in the social ecosystem that surrounds it, often highlighting societal tensions. This requires the accumulation of sufficient knowledge around that subject and then a critical mass of movement on that subject for the artwork to become relevant. There are three primary drivers for any piece of creative work: novelty, quality, and relevance.
Creativity plays several roles in society such as that of improvement, pushing us to imagine what the future can look like. It also plays the crucial role in self-expression for individuals that are a part of society. And finally, that of transformation of knowledge that helps to create innovations by combining existing tools and techniques in novel ways. The paper positions the creativity process as a selection-variation algorithm that helps us explore the space of possibilities, choosing to pursue some and ignore others which shapes societal structures.
Since intelligence and creativity tend to be quite intertwined, the authors provide a crude distinction between them classifying intelligence as a convergence methodology while creativity being a divergent methodology. Taking this on as a lens, we can now think of how we might want to evaluate the creative outputs from a machine. There are many criteria such as ideational fluency (amount of answers), originality (unusualness of answers), flexibility (variation in the concepts elaborated), and elaboration (precision and details of the answers) that are used to evaluate divergent thinking tasks and which can be ported here to evaluate creativity emerging from AI.
Given that AI systems are set up to optimize against certain targets, this poses inherent limits to the space of possibilities that it can explore. The solutions within that space can have infinite variations though. There are of course similar constraints on human creative output as well: through the social structures within they are present.
Intrinsic limits of AI for a self-contained creative agent¶
In the Wallas model of creativity that spans the stages of preparation, definition, incubation, illumination, and verification, we can see the role that AI plays in the preparation phase that includes information gathering. Since AI is able to process large amounts of information and surface potential connections between distant but related items, it can aid in the creative aspect of generating novelty. Another aspect is the ability of AI systems to transform data into different representations (the idea of representation learning) which can help human creators reimagine their own knowledge base in a different light sparking creativity along different axes using the same material. However, AI ends up being quite limited in the problem finding and illumination phases of the creative process, and that might just be limitations of current techniques; nonetheless, it provides us with a roadmap for where AI can be used effectively in the creative process.
Going back to the idea of novelty, looking at traditional machine learning approaches, outliers tend to be discarded as the system is hunting for dominant patterns that can be generalized. Given this inherent limitation, approaches such as those from reinforcement learning where there is an exploration-exploitation tradeoff might be more appropriate to deploy AI. But, this requires the specifications of rewards and success functions which are still limited by the ability and perspective of the human that can place artificial constraints on the AI system.
Coming up to the concept of relevance, since that is something that can be done only in the context of a social ecosystem, this becomes challenging for an AI system to do so in a self-contained manner.
Co-creativity as a way forward¶
So, perhaps rather than emphasizing and evaluating the outputs from such systems, the focus should be on the creative process and the interaction between the human and machine artists. Thinking back to how AI can help organize existing material in a different fashion, it can serve as a creativity facilitator. This comes with the caution that human limits transfer onto the limits of AI in the way that we define the problem and evaluate the outputs from the system, hence the focus on process rather than the output. Thus, we can view artificial creativity as an emergent phenomenon where the interaction between machine and human is what matters the most. Finally, a benefit of AI being able to process large, multivariate data is that it can expand our perception of creativity beyond the dimensions that we’re used to thinking in and move towards wholly novel forms of creative art.
Between the lines¶
The paper provides a well-rooted framework to reason about how we should think about AI systems in the context of creativity and how we might go about incorporating them into our creative workflows. For the most part, if we continue to push machines to emulate human creativity, we will end up with cheap imitations rather than expansions of the creative horizon. The benefits that machines bring are unique and beyond the abilities of humans; at the same time, humans will continue to be ones who can situate the work emerging from such processes to impart meaning to them. Hence, the focus in this work on the interaction and process much more so than the outputs itself is what is interesting. Certainly something for those in the intersection of AI and art to think more about.
What does this mean for Actionable AI Ethics?¶
The design of the system and its situational position in society is much more important than just the output itself. When thinking about the systems that we are making, thinking about the societal context is so crucial.
Questions that I am exploring¶
If you have answers to any of these questions, please tweet and let me know!
How do we better capture the interaction between machines and humans as we seek to evaluate the creative contributions of AI systems?
Potential further reading¶
A list of papers that I think might be interesting related to this paper.
Please note that this is a wish list of sorts and I haven’t read through the papers listed here unless specified otherwise (if I have read them, there will be a link from the entry to the page for that.)
I’ll write back here with interesting points that surface from the Twitter discussion.
If you have comments and would like to discuss more, please leave a tweet here.